Supercomputing Infrastructure Research Articles

The SCRUM-MONSTAR Cancer-Omics Ecosystem: Striving for a Quantum Leap in Precision Medicine.

The SCRUM-Japan MONSTAR-SCREEN consortium is a nationwide molecular profiling project employing artificial intelligence-driven multi-omics analyses for patients with advanced malignancies, aiming to develop novel therapeutics and diagnostics and deliver effective drugs to patients. Concurrently, studies assessing molecular residual disease-based precision medicine for resectable solid tumors, including CIRCULATE-Japan, are ongoing. The substantial data generated by these platforms are stored within a state-of-the-art supercomputing infrastructure, VAPOR CONE. Since 2015, our project has registered over 24,000 patients as of December 2023. Among 16,144 patients with advanced solid tumors enrolled in MONSTAR-SCREEN projects, 5.0% participated in matched clinical trials, demonstrating a 29.2% objective response rate and 14.8-month median survival (95% confidence interval, 13.4-16.3), for patients treated in the matched clinical trials. Notably, patients who received matched therapy demonstrated significantly prolonged overall survival compared with those who did not (hazard ratio 0.77; 95% confidence interval, 0.71-0.83).

There is no Open Science without infrastructure and science community

This paper focuses on the role of national-level infrastructure in promoting open science principles, with a specific emphasis on Slovenia. It presents the latest measures implemented within the country to facilitate open science practices.The first part of the paper explores the specific measures adopted in Slovenia to promote open science. It highlights the ongoing development of regulations to implement scientific research following open science principles. The Slovenian Scientific Research and Innovation Activities Act (ZZrID) and the Resolution on the Slovenian Scientific Research and Innovation Strategy 2030 (ReZrIS30) strongly emphasise open science and encompass various measures. The paper discusses the accessibility of research data addressed by the ZZrID and the Decree on the implementation of scientific research work in accordance with the principles of open science. Furthermore, it outlines upcoming actions in education and infrastructure solutions, as outlined in the Open Science Action Plan.The second part of the paper delves into the role of the Academic and Research Network of Slovenia (ARNES), the Slovenian national research and education network (NREN). ARNES has played a pivotal role in the construction of two new national data centres dedicated to the permanent storage of research data. The paper analyses the benefits of these national solutions, which enhance infrastructure accessibility, simplify researchers' work, and provide a more efficient and secure environment for research organisations to store their data. The development of data storage services and the construction of data centres take into consideration existing best practices and collaboration with researchers and research organisations. The solutions offered by the data centres will support existinguniversity repositories and align with the objectives outlined in the Open Science Action Plan. The integration of research funding and evaluation services into a centralised platform will be thoughtfully considered, taking into account the requirements of the research community. A key objective is to enable the implementation of services and infrastructure in accordance with OpenAIRE, where ARNES will play a pivotal role. As a mandated organisation in the European Open Science Cloud (EOSC) and being in touch with best practices in the EU area as an NREN, ARNES is well-positioned to contribute to this endeavour.The paper also highlights the broader objective of improving research quality, efficiency, and responsiveness within the national open science ecosystem. It discusses the establishment of the Slovenian Open Science Community (SSOZ) as part of the tripartite European Open Science Cloud (EOSC) event, making Slovenia one of the pioneering countries in the European Union. The SSOZ, along with Slovenian representatives in the EOSC, has already addressed challenges related to the permanent preservation of research data, which will be comprehensively discussed in relation to the new data centres and associated activities.Additionally, there is a growing demand for education on repository usage and infrastructure upgrades. The Slovenian Open Science Community serves as a unifying platform for all major universities and research institutions in Slovenia, fostering collaboration among its members. It encompasses participants from various initiatives, such as the NI4OS project, which aims to promote the European Open Science Cloud (EOSC) at a regional level. Furthermore, the community includes stakeholders from the Coalition for Advancing Research Assessment (CoARA) and similar European and institutional initiatives dedicated to advancing open science principles.Drawing from the experience of the case study, the paper critically highlights how the national infrastructure can simplify researchers' work. It emphasises the role of data centres in fostering sustainable and long-term data storage solutions aligned with the principles of open science. As an NREN, ARNES already manages public research infrastructure and offers solutions that facilitate research, such as data sharing and access to EuroHPC's supercomputing infrastructure. The establishment of new data centres by ARNES will significantly enhance the infrastructure for open science in Slovenia, providing primary conditions for implementing open science principles in research work. The implementation of planned professional training programs focused on utilising the infrastructure, along with workshops on open science organised by the Central Technical Library at the University of Ljubljana specifically for researchers in Slovenia, will serve as essential catalysts for enabling research activities that align with the principles of open science. With the infrastructure in place, these initiatives will empower researchers with the necessary knowledge and skills to embrace open science practices effectively.

CuHARM: A New GPU-accelerated GRMHD Code and Its Application to ADAF Disks

We introduce a new GPU-accelerated general relativistic magnetohydrodynamic code based on HARM, which we call cuHARM. The code is written in CUDA-C and uses OpenMP to parallelize multi-GPU setups. Our code allows us to run high-resolution simulations of accretion disks and the formation and structure of jets without a need for multinode supercomputer infrastructure. A 2563 simulation is well within the reach of an Nvidia DGX-V100 server, with the computation being about 10 times faster than if only the CPU is used. We use this code to examine several disk structures all in the SANE state. We find that (i) increasing the magnetic field in the SANE state does not affect the mass accretion rate; (ii) simultaneously increasing the disk size and the magnetic field, while keeping the ratio of energies fixed, leads to the destruction of the jet once the magnetic flux through the horizon decreases below a certain limit (this demonstrates that the existence of a jet is a function of the initial mass distribution, and not of the initial intensity of the magnetic field, since the magnetorotational instability dictates the evolution of the magnetic field); and (iii) the structure of the jet is a weak function of the adiabatic index of the gas, with relativistic gas tending to have a wider jet.

Национальная суперкомпьютерная инфраструктура для моделирования нооразвития

The paper discusses modern trends in the development of supercomputer technologies in the world and, particularly, in Russia. Russia’s place in the global supercomputer industry is determined; a close connection between the presence of supercomputers in a country and the level of its economic and technological development is indicated. It is shown that in Russia the available public supercomputers are used primarily to solve business problems, while educational and scientific problems are solved in much smaller volumes. This disrupts the connections in the “science – education – production” system and requires improvement. The most promising measure to correct the situation is the creation of a unified national supercomputer infrastructure. In this case, special attention should be paid to supercomputer modeling of socio-economic processes, in particular, scenario modeling of the civilizational transition from the current state of the economy and society through NIS.2 to noonomy.

G-SAIP: Graphical Sequence Alignment Through Parallel Programming in the Post-Genomic Era.

A common task in bioinformatics is to compare DNA sequences to identify similarities between organisms at the sequence level. An approach to such comparison is the dot-plots, a 2-dimensional graphical representation to analyze DNA or protein alignments. Dot-plots alignment software existed before the sequencing revolution, and now there is an ongoing limitation when dealing with large-size sequences, resulting in very long execution times. High-Performance Computing (HPC) techniques have been successfully used in many applications to reduce computing times, but so far, very few applications for graphical sequence alignment using HPC have been reported. Here, we present G-SAIP (Graphical Sequence Alignment in Parallel), a software capable of spawning multiple distributed processes on CPUs, over a supercomputing infrastructure to speed up the execution time for dot-plot generation up to 1.68× compared with other current fastest tools, improve the efficiency for comparative structural genomic analysis, phylogenetics because the benefits of pairwise alignments for comparison between genomes, repetitive structure identification, and assembly quality checking.

Optimization Strategies for Resource-Constrained Project Scheduling Problems in Underground Mining

Effective computational methods are important for practitioners and researchers working in strategic underground mine planning. We consider a class of problems that can be modeled as a resource-constrained project scheduling problem with optional activities; the objective maximizes net present value. We provide a computational review of math programming and constraint programming techniques for this problem, describe and implement novel problem-size reductions, and introduce an aggregated linear program that guides a list scheduling algorithm running over unaggregated instances. Practical, large-scale planning problems cannot be processed using standard optimization approaches. However, our strategies allow us to solve them to within about 5% of optimality in several hours, even for the most difficult instances. History: Accepted by Andrea Lodi, Area Editor for Design and Analysis of Algorithms—Discrete. Funding: This work was supported by Alford Mining Systems, the Centro de Modelamiento Matemático [Grants ACE210010 and FB21005], ANID-Chile [BASAL funds for center of excellence and FONDEF Grant ID19-10164], and the supercomputing infrastructure of the NLHPC [Grant ECM-02].

EESSI: A cross‐platform ready‐to‐use optimised scientific software stack

AbstractGetting scientific software installed correctly and ensuring it performs well has been a ubiquitous problem for several decades now, which is compounded currently by the changing landscape of computational science with the (re‐)emergence of different microprocessor families, and the expansion to additional scientific domains like artificial intelligence and next‐generation sequencing. The European Environment for Scientific Software Installations (EESSI) project aims to provide a ready‐to‐use stack of scientific software installations that can be leveraged easily on a variety of platforms, ranging from personal workstations to cloud environments and supercomputer infrastructure, without making compromises with respect to performance. In this article, we provide a detailed overview of the project, highlight potential use cases, and demonstrate that the performance of the provided scientific software installations can be competitive with system‐specific installations.

Supercomputer power consumption prediction using machine learning, nonlinear algorithms, and statistical methods

This work describes various methods of time series prediction. It illustrates the differences between machine learning methods, nonlinear algorithms, and statistical methods in their approach to prediction, and tries to explain in depth the principles of some of the most widely used representatives of these types of prediction methods. All of these methods are then tested on a time series from the real world: the course of power consumption of a supercomputer infrastructure. The reader is gradually acquainted with data analysis, preprocessing, the principle of the methods, and finally with the prediction itself. The main benefit of the work is the final comparison of the results of this testing in terms of the accuracy of the predictions, and the time needed to calculate them.

Towards Optimal Supercomputer Energy Consumption Forecasting Method

Accurate prediction methods are generally very computationally intensive, so they take a long time. Quick prediction methods, on the other hand, are not very accurate. Is it possible to design a prediction method that is both accurate and fast? In this paper, a new prediction method is proposed, based on the so-called random time-delay patterns, named the RTDP method. Using these random time-delay patterns, this method looks for the most important parts of the time series’ previous evolution, and uses them to predict its future development. When comparing the supercomputer infrastructure power consumption prediction with other commonly used prediction methods, this newly proposed RTDP method proved to be the most accurate and the second fastest.

On Abrupt Changes of the Approximate and Sample Entropy Values in Supercomputer Power Consumption

When calculating the Approximate Entropy (ApEn) and the Sample Entropy (SampEn) of a time series, representing the course of supercomputer power consumption, seemingly unexplained abrupt changes sometimes appear. These sudden changes occur when calculating the course of ApEn and SampEn values using a floating time window and can be manifested also in other discrete-valued time series, such as the development of prices on stock exchanges, time series from birth-death models, count-data time series, etc. These abrupt changes are not continuous and it is clear that they do not reflect real changes in the complexity degree of the analyzed time series. This can cause erroneous detection of a change in the degree of complexity of the time series, which can trigger a false alarm that something unusual is happening in the complex system being monitored (human brain, engine gearbox, financial market, supercomputer infrastructure, etc.). This work reveals in detail the mechanism of this phenomenon and also proposes measures to prevent its occurrence.

SECONIC: Towards multi-compartmental models for ultrasonic brain stimulation by intramembrane cavitation * * This work was carried out using the Supercomputer Infrastructure (STEVIN) at Ghent University, funded by Ghent University, the Flemish Supercomputer Center (VSC), the Hercules Foundation and the Flemish Government department EWI.

Objective. To design a computationally efficient model for ultrasonic neuromodulation (UNMOD) of morphologically realistic multi-compartmental neurons based on intramembrane cavitation. Approach. A Spatially Extended Neuronal Intramembrane Cavitation model that accurately predicts observed fast Charge Oscillations (SECONIC) is designed. A regular spiking cortical Hodgkin-Huxley type nanoscale neuron model of the bilayer sonophore and surrounding proteins is used. The accuracy and computational efficiency of SECONIC is compared with the Neuronal Intramembrane Cavitation Excitation (NICE) and multiScale Optimized model of Neuronal Intramembrane Cavitation (SONIC). Main results. Membrane charge redistribution between different compartments should be taken into account via fourier series analysis in an accurate multi-compartmental UNMOD-model. Approximating charge and voltage traces with the harmonic term and first two overtones results in reasonable goodness-of-fit, except for high ultrasonic pressure (adjusted R-squared ≥0.61). Taking into account the first eight overtones results in a very good fourier series fit (adjusted R-squared ≥0.96) up to 600 kPa. Next, the dependency of effective voltage and rate parameters on charge oscillations is investigated. The two-tone SECONIC-model is one to two orders of magnitude faster than the NICE-model and demonstrates accurate results for ultrasonic pressure up to 100 kPa. Significance. Up to now, the underlying mechanism of UNMOD is not well understood. Here, the extension of the bilayer sonophore model to spatially extended neurons via the design of a multi-compartmental UNMOD-model, will result in more detailed predictions that can be used to validate or falsify this tentative mechanism. Furthermore, a multi-compartmental model for UNMOD is required for neural engineering studies that couple finite difference time domain simulations with neuronal models. Here, we propose the SECONIC-model, extending the SONIC-model by taking into account charge redistribution between compartments.

New Fast ApEn and SampEn Entropy Algorithms Implementation and Their Application to Supercomputer Power Consumption.

Approximate Entropy and especially Sample Entropy are recently frequently used algorithms for calculating the measure of complexity of a time series. A lesser known fact is that there are also accelerated modifications of these two algorithms, namely Fast Approximate Entropy and Fast Sample Entropy. All these algorithms are effectively implemented in the R software package TSEntropies. This paper contains not only an explanation of all these algorithms, but also the principle of their acceleration. Furthermore, the paper contains a description of the functions of this software package and their parameters, as well as simple examples of using this software package to calculate these measures of complexity of an artificial time series and the time series of a complex real-world system represented by the course of supercomputer infrastructure power consumption. These time series were also used to test the speed of this package and to compare its speed with another R package pracma. The results show that TSEntropies is up to 100 times faster than pracma and another important result is that the computational times of the new Fast Approximate Entropy and Fast Sample Entropy algorithms are up to 500 times lower than the computational times of their original versions. At the very end of this paper, the possible use of this software package TSEntropies is proposed.

A Review of a Distributed High performance Computing Implementation

ABSTRACT This article presents the case of a regional government and academia working together to realize High-Performance Computing Wales, a shared supercomputing infrastructure. The initiative aimed to support economic development through creation of a distributed supercomputing capability, a skills development programme, and an initial programme of collaborative research and innovation. From the perspective of Smart Specialization, the study examines whether the economic development ambitions were achieved, and adds to the limited literature reviewing this aspect of High-Performance Computing Centers. Using principles of public sector investment guidelines, the study presents for Smart Specialization theory that such facilities can provide a general purpose technology. Further, from the perspective of practitioners, it highlights important steps in the appraisal, monitoring, and evaluation of such initiatives.

Hybrid multi-projection method using sparse approximate inverses on GPU clusters

The state-of-the-art supercomputing infrastructures are equipped with accelerators, such as graphics processing units (GPUs), that operate as coprocessors for each workstation of the distributed memory system. The multi-projection type methods are a class of algebraic domain decomposition methods based on semi-aggregation techniques. The multi-projection type methods have improved convergence behavior, as the number of subdomains increases, due to the corresponding augmentation of the semi-aggregated local linear systems with more coarse components, while the number of fine components is reduced. Moreover, limited amount of communications among the workstations is required by the proposed method. The utilization of the available GPUs allows an increase in the number of subdomains along with finer-grained parallelism, leading to improved performance. A load-balancing algorithm that ensures the concurrency of the computations on multicore processors and GPUs is proposed. Flexible parallel preconditioned Krylov subspace iterative methods enhanced with multi-projection type methods have been designed appropriately in order to have improved performance, compared to CPU-only or GPU-only executions, by exploiting the available CPUs and GPUs of the distributed memory system concurrently. The unsymmetric local linear systems are solved by the preconditioned Bi-Conjugate Gradient STABilized (BiCGSTAB) method enhanced with the modified generic factored approximate sparse inverse preconditioner, whereas the preconditioned conjugate gradient (CG) method along with the symmetric factored approximate sparse inverse preconditioner is used for the symmetric positive definite local coefficient matrices. Numerical results regarding the convergence behavior, the performance, and the scalability of the proposed method for several problems are given.

Machine learning for quality assessment of ground-based optical images of satellites

Astronomical images collected by ground-based telescopes suffer from degradation and perturbations attributed to atmospheric turbulence. We investigate the application of convolutional neural networks (CNNs) to ground-based satellite imaging to address two existing problems. First, the multiframe blind deconvolution (MFBD) algorithms that can extract well-resolved images from these degraded data frames are computationally expensive, requiring supercomputing infrastructure for relatively fast performance and currently cannot be done in real time. Because of this, it is difficult to optimize collection parameters to maximize the likelihood of producing a resolved image with MFBD. Second, the space-object National Imagery Interpretability Rating Scale (SNIIRS) allows human analysts to provide a quantitative score of image quality based on identification of target features. It is naturally difficult to automate this scoring process, not only because the scale is based on identifiable features but also because the images may be in an almost-resolved image quality regime that is difficult to handle for traditional computer vision techniques. For both applications, we present our results using CNNs on data collected at the Maui Space Surveillance Site as well as a new synthetic dataset we introduce containing over a million SNIIRS rated pairings of perturbed and pristine ground-based satellite images.

New frontiers in computing and data analysis – the European perspectives

ABSTRACTNowadays, Europe is heavily engaged in the evolution of a new generation of high-performance computing (HPC), and new perspectives are looking at the emerging computational era called exascale. The presence of converging, domain-independent interests on Big Data among different communities, from Information Communication Technologies (ITC) to Life Sciences, Physics, Astrophysics, Space Sciences and many other fields, as well as the development of new skills, is leading to the establishment of specific collaborations with the aim of exploiting knowledge, capacities and research and innovation potential from different communities. The new program of Europe (named EuroHPC) is encouraging the sharing of these experiences and the cooperation between different entities like scientific research centers, small and medium-sized enterprises (SMEs) and private industries, for joint actions and the development of new skills in the Big Data and artificial intelligence (AI) domains. This program has two main pillars: ‘Supercomputing and Data Infrastructure Programme’ to deliver two world-class precursors of exascale supercomputers on 2020 and ‘Research and Innovation Programme’ on HPC to support the development of European supercomputing technology with the co-design of European exascale machines and to foster applications, skills development and a wider use of the new generation of computing systems. The co-design of these new systems, considering the big amount of data they need to handle with increasingly complexity, will also drive the development of a new generation of artificial intelligence techniques and visual analytics systems to underpin more accurate investigations. This paper is an analysis of the evolution of computing from a European perspective, looking at new generation systems and data analysis techniques in the era of Big Data, illustrating ideas with examples developed within the context of a specific physics field.

Infrastructure of Supercomputing Technologies

Supercomputing technologies have a lot of aspects, and, speaking of their serious support at the state level, it is necessary to create and develop all elements of a supercomputing infrastructure, not focusing only on its individual components. In this paper, we discuss all basic elements of this infrastructure and illustrate the extreme demand for it in Russia, where only the list of tasks published by the Ministry of Education and Science of the Russian Federation contains more than 700 problems that require supercomputing resources for their solution. Many researchers have worked in this field; however, in this paper, we overview only the results achieved by some prominent Russian scientists from the Moscow State University, to which M.R. Shura-Bura devoted a significant part of his life.

Hyperparameter search in periodic vehicle routing problem

The large number of real-world applications have shown that the use of computational method for distribution process planning produces substantial savings. Many of these applications lead to problem generally known as Vehicle Routing Problem. The real-world applications are highly computationally demanding for larger instances. This article aims to show the possibilities and benefits of using hyperparameter search for solving the Periodic Vehicle Routing Problem for exhausted oil collection by execution on the supercomputing infrastructure using HyperLoom platform. HyperLoom is an open source platform for defining and executing scientific pipelines in a distributed environment. This experiment was run on the supercomputer Salomon operated by IT4Innovations.

A Big Data approach for multi-experiment data management

Data sharing among similar experiments is made difficult by the usage of ad hoc directory structures, data and metadata naming and by the large variety of access protocols. The Big Data paradigm provides the context to overcome the current heterogeneity problems. In this work, we present a study for a Global Storage Ecosystem designed to manage large and distributed datasets, in the context of physics experiments. The proposed environment is based on HTTP/WebDav protocols, together with modern data searching technologies, according to the Big Data paradigm. The main goal is to aggregate multiple storage areas and to simplify the operations of data retrieval, using Elasticsearch and Apache Lucene library. This platform offers to physicists an effective instrument to simplify the multi-experiment data analysis without knowing a priori the directory format or the data itself. As a proof of concept, we realised a prototype over the ReCaS Supercomputing infrastructure in Napoli.

Making Supercomputing Available to All Cuban Researchers

Scientists in Cuba must get up to date with the latest trends and best practices in scientific computing. As a consequence of the socioeconomic and geopolitical situation in and around Cuba, a lack of financial means restricts Cuban researchers access to the supercomputing infrastructures, techniques, and tools that are urgently needed to support the increasing use of computational tools and efficient data processing in different areas, especially those related to public health.